OpenAI has finally unveiled its first custom-built processor, developed in collaboration with Broadcom. Named Jalapeño, the chip is specifically designed for inference, the process of running pre-trained AI models in response to user commands. The announcement today marks a strategic step to reduce dependence on Nvidia's dominant GPUs.
Jalapeño: a chip tailored for inference
The new processor was designed using OpenAI's own AI models to optimize the design. According to the company, early tests show significantly better performance-per-watt than current state-of-the-art alternatives. This is crucial because inference represents the most costly operational phase for generative AI, and any cost reduction can have a huge impact on company finances.
Energy performance and independence from Nvidia
With Jalapeño, OpenAI follows in the footsteps of Google and Amazon, which have already developed custom AI chips. The goal is twofold: increase energy efficiency and reduce dependence on a single supplier. The chip is specifically optimized for real-time coding models, but more intensive workloads like pre-training will likely still run on Nvidia hardware. However, even small reductions in inference costs can significantly improve margins.
Sponsored Protocol
OpenAI's full-stack vision
Greg Brockman, OpenAI's president, explained the approach in an internal podcast: "We have a deep understanding of the workload. We're looking for specific underserved workloads and building something that accelerates what's possible." This fits into a broader strategy: OpenAI not only develops cutting-edge models and products like Codex, but also the underlying infrastructure, including chips, kernels, memory systems, networking, and scheduling. By operating across the entire tech stack, every layer can be optimized for the same goal: making models faster, more reliable, and more affordable. For an example of how multi-agent frameworks are evolving, see Anthropic launches Claude Tag, a persistent AI agent that works in Slack. Additionally, frameworks like AutoGen and CrewAI show how multi-agent complexity requires increasingly specialized infrastructure.
Sponsored Protocol
Economic impact and future of AI infrastructure
Optimizing inference is a crucial factor in AI economics. With custom chips like Jalapeño, OpenAI aims to reduce operational costs and accelerate the adoption of AI agents. The company also announced continued investment in data centers and deployment systems. For more on AI accelerators, see the Wikipedia entry. With this move, OpenAI demonstrates that the future of AI also involves controlling the hardware.
Source: https://techcrunch.com/2026/06/24/openai-unveils-its-first-custom-chip-built-by-broadcom
